Analyze tool

[Pippen]

Hi.

I'm looking for an computer analyze tool for my games like that you load the game in and the computer calculates for the whole game and shows you which move it thinks were suboptimal etc. I know it from Chess where it was a pretty cool feature. I'm 1k, so even if computers aren't that strong I could get some valueable info I guess. Is there any programm with such an feature? I was thinking of MFOG. It is said to have a score graph where it calculates and shows your winning percentage based on the whole games? Maybe if one can set up like 100.000 trial games (MFOG uses MC) one could get a pretty accurate pic of one's game. Thoughts?

[Phelan]

The only one I remember having something like that is Many Faces, like you said. I don't know how it calculates it, though.

[smartgo]

SmartGo for Windows can analyze your game (Tools > Analyze Game) and show a score graph; you can see an example at http://www.smartgo.com/games.htm. However, while it's useful for weaker players, in games at your level it's more likely to highlight mistakes in the computer algorithm than mistakes in your game.

[Rémi]

How about that:
http://www.grappa.univ-lille3.fr/~coulo ... index.html

It is still an experimental feature of Crazy Stone. In order to test it, I can offer a few Crazy Analysis to readers of this forum. Just give a link to a sgf file. I'll analyze the first 10 games with Crazy Stone.

Rémi

[perceval]

Hello Remi,
here is a 6k game, i would like to take you up on your kind offer forcomputer analysis
http://files.gokgs.com/games/2012/1/16/AMak-tictac.sgf

Not really interesting or anything, i know there was some big reading errors, but i am really curious to use your tool. there was a big fight going on i am curious to see how a computer would have handled it. Plus i guess it might be interesting also to see how the analysis handle a weaker game maybe ? probaly correction at each move and huge delta. i am curious to see move labeled with a delta of +0.3. there was no such move on the tromp game you linked.

i ll take the opportunity to express my admiration for crazystone which launch the incredible surge in go computer strength of this day . its not every day that you can really make such a breakthrough in a field. if my days were 36 hours ill definitly occuppy 5 or 6 to make a gobot.
Do you feel that at some point your level in go will be a blocking point for dev ? your creature is already 9 stones stronger than you ...

[Rémi]

Hi perceval,

http://www.grappa.univ-lille3.fr/~coulo ... index.html

A bit more delta than in the tromp game, indeed

Rémi

[hyperpape]

I'm curious to see how it performs on really high level games that were close. One wouldn't expect it to necessarily find a lot of mistakes, but it might be interesting what it thought winning percentages were.

[Uberdude]

Hi Remi,

Thanks for the offer. I would be extremely interested to see Crazy Stones's analysis of this game:

http://www.online-go.com/games/sgf/286/ ... tality.sgf

We discussed it a bit in the thread viewtopic.php?f=4&t=5277 and around move 40 we couldn't decide what was the best. I think Monte Carlo bots are particularly good at these sort of centre moves and unconstrained by our human way of thinking.

Thanks.

EDIT: I only just clicked on the move hyperlinks: it's great the way it shows prospective territory, the choice of moves and the expected sequences too.

[Cassandra]

Hello, Remi,

to be honest, it's no game, but I would be really interested in having an analysis by Crazy Stone, if possible.

TRMDPE2011_4CrazyStone.sgf is what we currently think to be the solution of Igo Hatsuyoron 120, but may be handicapped by the existence of the Hanezeki in the lower right, which might be beyond a program's capabilities.

VAR1929_4CrazyStone.sgf perhaps might be preferable (it's one of numerous variations), because the Hanezeki will be turned into a real Seki by some additional moves, to prevent the program from thinking about capturing the Hanezeki's tail.

So far, we got some valuable insights from using the known features of Many Faces.

Thanks a lot
Thomas

[Rémi]

http://www.grappa.univ-lille3.fr/~coulo ... index.html

The level of play of this game is much higher. Maybe a bit too high for Crazy Stone. The analysis is still in progress. I will let it search for a while (if you refresh the page, you'll see the number of playouts increases from time to time).

Most striking is that Crazy Stone believes the final position is balanced.

Rémi

[Rémi]

Hi Thomas,

I'll try but later. I'll be busy in the days to come. Probably I'll run the analysis during the week-end.

Your problem looks intimidating. Crazy Stone is more likely to produce useful results when analyzing beginner games. We'll see.

Rémi

[Uberdude]

Thanks! I will study this...

The level of play of this game is much higher. Maybe a bit too high for Crazy Stone.

I think if we play on KGS real-time Crazy Stone and Zen are probably stronger than me now, but on OGS probably I am stronger. So if it says a move is better I won't believe it automatically, but will seriously consider it.

Most striking is that Crazy Stone believes the final position is balanced.

Hehe, that is certainly a human weakness to resign after such a psychologically devastating move as my cut there which was a swing of over 20 points. But as I was quite a bit behind before I am not leading so much after it. I played out a yose and got black ahead 5-10 points. Certainly close enough that if I play poor yose (unlikely on OGS) and white good yose white could still win.

[Cassandra]

Hi Thomas,
I'll try but later. I'll be busy in the days to come. Probably I'll run the analysis during the week-end.

Your problem looks intimidating. Crazy Stone is more likely to produce useful results when analyzing beginner games. We'll see.

Rémi

Dear Remi,

Please keep your time.
My "attack" was just an idea, because you were so kind to post here.

May be that it is really much too early, when you are starting to implement usuful tools for analysis.

Related to this problem, Many Faces had shown a dramatic shift in quality of the results with version 12.21. In most cases (especially when it "feels" slightly behind) Many Faces gives multiple candidate moves. For a 1 Kyu like me, it was not so decisive that not all of these candidates were correct in every case, but a program has the big advantage not to overlook a valid potential candidate move, only because it did not work several times / variations before.

[perceval]

Hi perceval,

http://www.grappa.univ-lille3.fr/~coulo ... index.html

A bit more delta than in the tromp game, indeed

Rémi

Thanks a lot. The sorting of the move by delta is an interesting view to spot the blunders biggest delta was 0.23 . I suggested this game because I knew there was quite a lot of swings in winning rate, and errors and indeed there was
I feel that for games at this level this is an incredible tool to detect tactical blunders, but also endgame errors: there was quite some errors there it seems. the territory view is also great.
I am surprised a bit that even in the fuseki crazy seems to propose normal moves : we had some games with fuego on this forum and the fuseki was really weird.

Tonight I'll try if time allow to add crazy'z suggestions to an sgf and post this game in the analysis forum to confront crazystone's analysis to our human experts here. I am curious what they'll have to say. In particular around the huge ko around move 134 crazy seemed a bit reluctant to start it which i found weird

@pippen: sorry this thread seems to derail a bit